Hong Kong has delayed proposed legislation addressing the use of copyrighted material to train artificial intelligence systems, choosing to develop operational guidelines before moving ahead with statutory amendments. The decision highlights the difficulty of balancing copyright protection, AI development and the interests of the jurisdiction’s creative and technology sectors.

The proposal was contained in the Copyright (Amendment) Bill 2025. It would have introduced an “opt-out” mechanism for text and data mining, allowing copyright owners to signal that their works should not be used for certain AI-related purposes. The bill was originally expected to be submitted to the Legislative Council in the first half of 2025, but it was later removed from the legislative agenda.

The latest development suggests that the government is not abandoning copyright reform, but is proceeding more cautiously. Instead of immediately changing the Copyright Ordinance, authorities are now drafting a code of practice on copyright issues and the application of legal principles relating to generative AI, as well as guidelines explaining existing legal principles and their application.

The choice of an opt-out system has been controversial. Critics have argued that it places a substantial burden on rights holders, who would need to monitor whether their material had been used and then approach AI companies if they wanted to exclude it. As reported by the South China Morning Post, some critics said the process would “involve a lot of effort” for copyright holders. They favoured an opt-in model requiring permission before protected works could be used.

The distinction is commercially significant. Under an opt-in system, AI developers would generally need to obtain consent before training on protected works. Under an opt-out system, use may be permitted unless the owner takes steps to reserve its rights. Each model distributes compliance costs differently and creates different incentives for licensing, collective management and technical filtering.

For media companies, publishers, music businesses, visual artists and other creators, uncertainty remains a central concern. Many rights holders want clarity over whether data scraping, indexing, model training and output generation infringe copyright. They also want workable methods for identifying the use of their content and negotiating compensation.

AI developers, meanwhile, are likely to prefer rules that are predictable, technologically neutral and compatible with cross-border data practices. Training datasets often contain material from multiple jurisdictions, and companies may find it difficult to comply with different national standards. Hong Kong’s decision to issue guidance first could therefore provide flexibility, but it may also prolong uncertainty if the guidance does not answer questions about liability, licensing and enforcement.

Hong Kong already has a broader AI governance framework. The Digital Policy Office’s Generative Artificial Intelligence Technical and Application Guideline recommends transparency measures such as watermarks, labels, metadata and digital signatures where generative AI has been involved in content generation or decision-making. The government has also indicated that existing laws should be reviewed for gaps before deciding whether bespoke legislation or administrative measures are needed.

The copyright delay comes as other jurisdictions across the Asia-Pacific region examine similar questions. Singapore has opened a consultation on AI’s impact on copyright and patents, while Japan and South Korea have already put broader AI governance laws into effect. The resulting regional landscape is likely to be fragmented, with differences over training exceptions, licensing, authorship, inventorship and transparency.

For Hong Kong businesses, the immediate practical response should be risk management rather than waiting for legislation. AI users should maintain records of the sources of copyright works used for AI training, review licences, require contractual assurances from vendors and establish procedures for responding to takedown or rights-holder complaints. Creative businesses should consider how their works can be identified and whether contractual or technical reservation tools are available.

The decision to delay legislation may ultimately prove useful if it allows the government to test industry guidance and gather evidence before fixing a particular legal model. But it also means the current uncertainty will continue. The central policy question remains unresolved: Should Hong Kong prioritize permission-based access to creative works, or seek a more flexible framework that supports AI development while relying on transparency, licensing and enforcement after the event?

- Asia IP